q2-ANCOM versus R-ANCOM-2

Hello Qiimers!
Recently, I noticed some weird outputs from the q2-ANCOM, which encouraged me to test the ANCOM-2 in R using the original script from the authors of ANCOM. By comparison, I realized that the output from q2-ANCOM may not be reliable but ANCOM-2 is fully reproducible and robust. I happened to communicate off late with the senior author of ANCOM and received the link to download the latest ANCOM-2 R package. I am told that the volcano plot will be added pretty soon to the package. Hope this will help those who are using ANCOM for predicting the differential features.

Cheers!
BS

1 Like

Hi @bsen2018,
Thanks for the update and link!

Just in case your weird outputs were related to low W values wrongly being identified as significant, this thread may shed some light on that.

1 Like

@Mehrbod_Estaki, Yes, I did read through most of the threads related to ancom and consequently thought of sharing my experience of comparative evaluation with the two scripts.

1 Like

Hi @bsen2018,
It sounds like you are using the ANCOM 2.0 test data for your comparison? That makes sense since you are replicating the results with the ANCOM 2.0 tutorial.

It also probably makes sense that you are seeing different results with q2-composition, which uses an implementation of ANCOM 1.0, not ANCOM 2.0 (which uses different tests under the hood). So different results between these implementations are not necessarily surprising… they may even be expected!

Hello @Nicholas_Bokulich,
That being said, are there any plans by the q2 developers to upgrade the q2-composition plugin to the newer version of ancom, i.e., ANCOM-v2.1, in the near future?

Yes, I can’t guarantee an ETA, but it is on our radar.